BG05 Artificial intelligence prediction of splice site mutation consequences in junctional epidermolysis bullosa: signposts to severity

Abstract

Abstract Over one-fifth of disease-causing mutations in junctional epidermolysis bullosa (JEB) are splice site variants. These affect pre-mRNA splicing through the activation of cryptic splice sites and/or exon skipping, resulting in altered mRNA and translated proteins. Prediction of the precise location of activated cryptic splice sites is vital, as these can potentially alter the reading frame of the resulting transcript. Out-of-frame transcripts contain premature termination complexes (PTCs), which generally result in nonfunctional protein and often correlate with severe JEB in conjunction with other PTC-introducing mutations. SpliceAI is a 32-layer convolutional neural network that examines 10 000 nucleotides of flanking sequence around a variant of interest to predict its effects on RNA splicing. It was used to predict splice site mutation consequences in five patients with JEB with LAMB3 mutations. Genotypes, predicted transcripts and JEB subtypes are summarized in Table 1. SpliceAI predicted PTCs in cases 1 and 2, where transcripts were either out of frame or novel PTCs were introduced following aberrant splicing. An in-frame transcript with no PTC was predicted for case 3, which may explain this individual’s mild phenotype. Reverse transcriptase polymerase chain reaction data were reviewed for cases 4 and 5, and two predicted transcripts were confirmed. Two additional in-frame transcripts not predicted by SpliceAI were found to be produced in low levels. It is likely that these in-frame transcripts generated partially functional protein, alleviating JEB severity to a degree. This study highlights the importance of examining RNA transcripts for accurate genotype–phenotype correlation. In silico tools such as SpliceAI show potential in elucidating mechanisms of severity. They may facilitate selection of cases for further laboratory investigation and can guide which regions of the gene or transcript to sequence for confirmatory analyses. Where RNA analyses were available, SpliceAI correctly predicted two of the resultant transcripts. Further validation of this tool is required with additional cases in this rare disease.Table 1Genotypes, predicted transcripts and JEB subtypes found with SpliceAICaseZygosityMutation (LAMB3)Exon skipping predictionCryptic splice site activation predictionPredicted outcome commentsJEB subtype1Homozygousc.2701+1G>AOOF exon 18 skipping20 nt excluded from exon 18Both outcomes OOFSevere2Heterozygousc.565-2A>G, c.2914C>T (p.R972X)OOF exon 7 skipping117 nt included from intron 6 (including TGA PTC at c.565-60)In-frame cryptic activation with novel PTC introducedSevereOOF exon skipping3Heterozygousc.1705C>T (p.R569X), c.943+2T>COOF exon 9 skipping27 nt included from intron 9In-frame cryptic activationIntermediateOOF exon skipping4Homozygousc.298+5G>COOF exon 4 skipping64 nt included from intron 4Both outcomes OOFIntermediate5Heterozygousc.3119G>A (p.W1040X), c.629-12T>APredicted to be unlikely10 nt included from intron 7OOF cryptic activationIntermediatent, nucleotides; OOF, out of frame.